Locate the added voice
Listen to generated source footage before the edited output. Subtitles alone cannot reveal a filler audible in the track, and automatic transcription may misrecognize correct speech.
| Audible problem | First check |
|---|---|
| Source contains added narration or dialogue | Its complete sound timeline and the scene’s established sounds. |
| A filler precedes the approved line | Whether speech begins with the first original word and whether fillers were permitted. |
| One line occurs twice | Repeated text in the prompt or adjacent shots both starting from its beginning. |
| Source is correct; output adds speech | Edit tracks, narration, duplicated segments and placement. |
This page addresses unapproved voices. Spoken-language errors and wrong-mouth animation have separate guides. Do not regenerate correct source merely because an error appears in the final edit.
Write permitted sounds and confirmed exclusions
For sound generated with picture, Drama Skills’ motion recipe lists all dialogue, narration, off-screen speech, needed nonverbal events and effects. Exact lines appear once in the copyable shot text. Outside those events, retain declared ambience or silence; the visual section refers to sound events without repeating their words.
Each voice event needs speaker, source, language, exact line, and beginning/end. Include sighs, laughter, breaths or overlap when the scene chooses them rather than prohibiting them universally.
“No dialogue” followed by a speaking action contradicts itself. “No voiceover” alone also fails to identify required speech and effects. Summarize the actual layers at the end and consolidate only the creator’s confirmed exclusions.
Complete example one: keep the clerk’s line
Original paper scene: printing-shop clerk Meng Ru hands one sample booklet to Gu Yan across a counter. Meng’s visible dialogue is the Mandarin line shown below. Gu does not speak. Established sound consists only of a continuous low fan, that line and slight paper friction during receipt. The creator excludes narration, new lines, extra fillers, repeats and music. Duration still needs actual speech, handoff and ending checks; no footage or listening result exists.
This candidate can be adapted to established character and location facts. It claims no existing reference image or recording:
Fixed two-person medium shot. Meng Ru is behind the counter, Gu Yan in front, and Meng holds one sample booklet. Start in these positions with the established low fan ambience.
Meng places the booklet at the counter’s handoff position while Gu waits. After placing it, sound event A begins: Meng Ru, visible dialogue, spoken Mandarin, verbatim “这本先给你核对。” Speech begins with its first word and ends with its last. Gu listens without vocalizing, lips closed. The visual segment does not repeat that line.
After event A finishes, Gu takes the booklet and Meng releases it. Keep slight paper friction from the handoff and continuous fan ambience. End with the booklet in Gu’s hands and Meng’s hands empty; neither vocalizes again. Sound layers are this visible line, fan ambience and the stated paper sound. Add no narration, new lines, extra fillers, repeats or music.
The line means “This copy is for you to check first.” This explanatory translation is outside the generation instruction and is not a second spoken line. An English prompt body can retain the Mandarin original.
The line is dialogue, not narration; relabeling it to avoid mouth animation changes the scene. Do not append narration announcing successful receipt. Receiving the booklet does not mean Gu has checked its contents.
Complete example two: a separately chosen no-dialogue version
If the creator separately adopts a script without dialogue, the same transfer can use ambience and action sound alone. Choose that version before production rather than deleting required speech while repairing added narration:
Fixed two-person medium shot. Meng holds one sample booklet behind the counter; Gu waits in front. Throughout, retain only the continuous low fan and slight paper friction during receipt. Neither person vocalizes; both keep lips closed.
Meng places the booklet at the handoff position. Gu then takes it and Meng releases it. Paper sound follows this actual handoff; before and after it, fan ambience continues without filling the gap with another sound. End with the booklet in Gu’s hands and Meng’s hands empty, holding that state. No dialogue, narration, off-screen speech, other voices or music; retain the declared fan and paper sounds.
No dialogue is different from a muted film. This version has no sigh or laughter. Where a script needs those events, state who produces them and when instead of copying this exclusion set. Prompt language, actual spoken language and subtitle text are separate decisions.
Listen after generation and repair the right layer
Listen from beginning to end for added prefixes, endings, repeats or changed speakers; then inspect whether Gu speaks along during receipt. Compare actual audio rather than correct-looking subtitles. Locate the error in source or post-production and change the affected segment or track.
If unwanted narration shares a track with required dialogue, muting removes both. For regeneration, separate dubbing or post-production, confirm that the actual tool can perform the chosen treatment, then listen to its output. The suite’s edit sound notes do not automatically execute muting, bridges or track replacement.
Wording does not prove successful exclusion. Preserve the full required line, correct speaker, needed effects and intended ending as well as removing additions.
FAQ
Is “no voiceover” enough?
Also specify permitted dialogue, effects, ambience and intervals. State what to retain in a complete timeline, then listen to the result.
Can muting the source solve it?
Only when the creator chooses to discard the source track and has an adopted replacement. Whole-track muting also removes dialogue or effects that must remain.
Do it with the skill
Use $short-drama-video-prompts with the scene script, exact lines, speakers, source types, languages, ambience/effects and observed additions. Ask for one occurrence of each line, established sound between events and only confirmed exclusions. Preserve visuals and ending; a prompt revision is not a repaired film.
Read the method: Complete sound timeline and exact lines · Sound source, speaker and shot continuity · Actual sound treatment and listening
About Drama SkillsThe short-drama-video-prompts skill on GitHub